Automated training data quality process

1. A system for updating training data, comprising:

an interface configured to receive a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

a processor; and

a memory coupled with the processor, wherein the memory is configured to provide the processor with instructions which when executed cause the processor to:

determine a set of training data and a set of test data from the set of vehicle data, comprising to:

determine a filtered set of training data from the set of vehicle data; and

determine the set of training data from the filtered set of training data;

train a model with the set of training data, wherein the set of training data comprises a) the filtered set of training data decimated to a fixed number of frames, b) the filtered set of training data cropped to a driver location portion of an image, or c) the filtered set of training data normalized to a data value range;

determine a set of predicted labels for the set of vehicle data using the model;

identify a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determine an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.

2. The system of claim 1 , wherein the processor is further configured to retrain the model using the updated set of training data.

3. The system of claim 1 , wherein identifying a potential mislabeled data comprises determining whether a predicted label probability is greater than a first confidence level and a predicted label is different from an assigned label.

4. The system of claim 1 , wherein identifying a potential mislabeled data comprises determining whether a predicted label probability is less than a second confidence level and a predicted label is identical to an assigned label.

5. The system of claim 1 , wherein the filtered set of training data comprises the set of training data filtered based at least in part on whether image quality is above an image quality threshold.

6. The system of claim 5 , wherein the image quality threshold comprises an uncompressed image quality threshold.

7. The system of claim 1 , wherein the filtered set of training data comprises the set of training data filtered based at least in part on whether a driver seating position corresponds to right-side road driving.

8. The system of claim 1 , wherein the filtered set of training data comprises the set of training data filtered based at least in part on whether date of training data is within a date range.

9. The system of claim 1 , wherein the fixed number is set to using a specific number of frames per second.

10. The system of claim 1 , wherein the fixed number is set to a specific number of reference I-frames.

11. The system of claim 1 , wherein the set of training data comprises the filtered set of training data resized to a model input image size.

12. The system of claim 1 , wherein training the model includes performing a test of the model using the set of test data.

13. A system for updating training data, comprising:

an interface configured to receive a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

a processor; and

a memory coupled with the processor, wherein the memory is configured to provide the processor with instructions which when executed cause the processor to:

determine a set of training data and a set of test data from the set of vehicle data, comprising to:

determine a filtered set of training data from the set of vehicle data; and

determine the set of training data from the filtered set of training data;

train a model with the set of training data, wherein training the model ends when a test shows a performance improvement less than a threshold performance improvement compared to a last iteration test;

determine a set of predicted labels for the set of vehicle data using the model;

identify a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determine an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.

14. A method for updating training data, comprising:

receiving a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

determining, using a processor, a set of training data and a set of test data from the set of vehicle data, comprising:

determining a filtered set of training data from the set of vehicle data; and

determining the set of training data from the filtered set of training data;

training a model with the set of training data, wherein the set of training data comprises a) the filtered set of training data decimated to a fixed number of frames, b) the filtered set of training data cropped to a driver location portion of an image, or c) the filtered set of training data normalized to a data value range;

determining a set of predicted labels for the set of vehicle data using the model;

identifying a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determining an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.

15. A computer program product for video processing, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

receiving a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

determining, using a processor, a set of training data and a set of test data from the set of vehicle data, comprising:

determining a filtered set of training data from the set of vehicle data; and

determining the set of training data from the filtered set of training data;

training a model with the set of training data, wherein the set of training data comprises a) the filtered set of training data decimated to a fixed number of frames, b) the filtered set of training data cropped to a driver location portion of an image, or c) the filtered set of training data normalized to a data value range;

determining a set of predicted labels for the set of vehicle data using the model;

identifying a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determining an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.

16. A method for updating training data, comprising:

receiving a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

determining, using a processor, a set of training data and a set of test data from the set of vehicle data, comprising:

determining a filtered set of training data from the set of vehicle data; and

determining the set of training data from the filtered set of training data;

training a model with the set of training data, wherein training the model ends when a test shows a performance improvement less than a threshold performance improvement compared to a last iteration test;

determining a set of predicted labels for the set of vehicle data using the model;

identifying a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determining an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.

17. A computer program product for video processing, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:

receiving a set of vehicle data, wherein the set of vehicle data includes images and assigned labels associated with the images;

determining, using a processor, a set of training data and a set of test data from the set of vehicle data, comprising:

determining a filtered set of training data from the set of vehicle data; and

determining the set of training data from the filtered set of training data;

training a model with the set of training data, wherein training the model ends when a test shows a performance improvement less than a threshold performance improvement compared to a last iteration test;

determining a set of predicted labels for the set of vehicle data using the model;

identifying a set of potential mislabeled data using the set of predicted labels and the assigned labels; and

determining an updated set of training data by relabeling the set of potential mislabeled data and replacing the set of potential mislabeled data with a relabeled set of data.